Phonological Representations of Standard German
3.1 Standard German Stops in Phonology
Phonologically, laryngeal contrasts like the German two-way opposition be-tween tense and lax stops use to be represented by means of distinctive fea-tures. Since the German laryngeal distinction pertains to obstruents only, sonorants and vowels do not receive a specification for laryngeal features at all. For German stops, a classical textbook representation, as found, for example, in Hall (2000:107), suggests a featural specification as shown in figure 3.1.2 However, the representation of laryngeal contrasts in obstruents has been much debated by phonologists for more than half a century. One main aspect in this still continuing debate pertains to the finding of cross-linguistic differences in the phonetic realisations of tense/lax oppositions in two-way contrasting stop systems. As discussed in chapter 2, there are some
1Note that ‘pre-phonological’ is meant to refer to infants who have not yet entered a phase in which they start to systematically learn words and who have not yet begun to establish a mental lexicon. ‘Phonological’ in this connection is not intended to imply that children have mature, adult-like phonological representations.
2Hall uses the German equivalents [+stimmhaft] and [−stimmhaft] instead of the En-glish labelling [+voice] and [−voice].
3.1 Standard German Stops in Phonology 45 languages, including German, in which the tense/lax contrast is not one of real voicing in the sense of vocal fold vibration but rather a contrast of voiceless unaspirated versus voiceless aspirated stops. This fact as well as different phonological processes in languages with different laryngeal two-way contrasts raised doubts on whether a feature [voice] is an appropriate specification to capture the contrast in these languages.
Figure 3.1:Classical phonological representation of the German stop contrast with a binary feature [±voice].
3.1.1 Laryngeal Features
Various proposals have been made with regard to laryngeal features. Re-garding mono- or bivalence of features, Lombardi (1991, 1995b), for example, advances the view that [voice] is a privative feature and that voicelessness is never represented by [−voice]; it is rather the default realisation of obstru-ents that lack the laryngeal node. Wetzels & Mascar´o (2001), in contrast, claim that laryngeal properties must be phonologically specified by a binary feature [±voice] in order to be able to capture phonological processes like voice assimilation and devoicing cross-linguistically. Instead of a feature [voice] – be it privative or binary – other features have been suggested to represent laryngeal contrasts.
Based on a neat phonetic description of the operations at work in the glottis during speech production and the acoustic effects of various config-urations of the larynx and the vocal folds, Halle & Stevens (1971) propose four binary features for the classification of phonetic segments: [±spread glottis] and [±constricted glottis], referring to the degree of glottal opening, and [±stiff vocal folds] and [±slack vocal folds], referring to the status of vo-cal fold tension. Grounded on the physiologivo-cal insight by Kim (1970) that the glottis is wide apart during the articulation of aspirated stops, Halle &
Stevens (1971) show that glottal aperture is not responsible for voicing as has been presumed before. Instead, it is connected with aspiration. The wider the glottal opening at the moment of stop release, the greater the amount of aspiration (i.e., ACT). Voicing, on the other hand, is influenced
by vocal fold tension. In voiceless stops, the vocal folds are tensed or ‘stiff’, inhibiting vocal fold vibration. Stops that display closure voicing are pro-duced with slack vocal folds, allowing for glottal vibration during the stop’s closure. Halle & Stevens (1971) suggest that these features should replace traditional features like voicing, aspiration, glottalisation and vowel pitch.
Table 3.1 provides their featural description put forward for the phonemes as they surface in the different two-way laryngeal contrasts mentioned above.3
Feature [b, d, g] [p, t, k] [ph, th, kh]
spread glottis – – +
constricted glottis – – –
stiff vocal folds – + +
slack vocal folds + – –
Table 3.1: Featural representation of voiced, plain voiceless, and voiceless aspirated stops as given in Halle & Stevens (1971:51).
Halle & Stevens’s (1971) laryngeal feature framework has contributed much to the debate on laryngeal features and the question of specification. While [spread glottis] (or sometimes [aspiration], e.g., Lombardi 1991, 1995b) and [constricted glottis] became widely accepted features to specify aspirates on the one hand and ejectives and implosives on the other hand (e.g., Lahiri
& Reetz 2010), the features [stiff vocal folds] and [slack vocal folds] were only adopted by some (e.g., Halle 2003; Keyser & Stevens 2006). Wiese (2000) notes the features referring to vocal fold tension but decides to stick to the more common [±voice]. Moreover, the attention to the pho-netic implementation of stop contrasts brought forward other suggestions.
Some accounts prefer to specify the tense stop series positively (i.e., not by a ‘devoicing feature’ [−voice] but instead by a feature referring to the voiceless/fortis/aspirated quality of the tense category), proposing several other features. For instance, Kohler (1984) suggests to introduce a feature [±fortis], Jessen (1998) uses [tense], and Iverson & Salmons (1995, 1999) as well as Jessen & Ringen (2002) prefer [spread glottis]. A further motiva-tion to replace the tradimotiva-tional [voice] feature lies in the advantage of having
3Note that there is no role for Halle & Stevens’s (1971) laryngeal feature [±constricted glottis] in the present selection of prevoiced, voiceless unaspirated and voiceless aspirated stops. The feature is essential to cover implosives and ejectives. Both groups of obstruents are irrelevant in the two-way contrasts examined in the present work.
3.1 Standard German Stops in Phonology 47 identical features for both consonants and vowels. Halle (2003) suggests to replace the feature [±voice] by the feature [±stiff vocal folds]. The latter feature may be used to determine the laryngeal status of obstruents and at the same time it refers to pitch in vowels. Consequently, phonological processes concerning both vowels and consonants can be described as sim-ple feature spreads between the two classes of sounds. In this way, Halle (2003) explains the phenomenon of fricative voicing after unaccented vowels in Indo-European known as Verner’s Law as a simple assimilation process where the low-pitched vowel passes the feature [−stiff vocal folds] onto the following fricative. Similarly, in order to relate to both consonants and vow-els, Jessen (1998:257) suggests to use [tense] as an ‘articulator-free’ feature (i.e., not dominated by the laryngeal node in the feature hierarchy).
3.1.1.1 Two Hypotheses for Feature Specification
Since the phonetic implementation of tense/lax contrasts in stops as well as phonological processes pertaining to these stops vary across languages, the question of an appropriate feature specification for laryngeal contrasts goes beyond terminological issues. As mentioned already in chapter 2, languages with a two-way laryngeal stop contrast can be assigned to two different groups. Some languages make a contrast between prevoiced and voiceless unaspirated stops (voicing lead vs. short voicing lag), whereas other lan-guages oppose voiceless unaspirated to voiceless aspirated stops (short voic-ing lag vs. long voicvoic-ing lag; also cf. Lisker & Abramson 1964). The latter pattern is found in most Germanic languages including German and En-glish. The Romance and the Slavic languages and also the West Germanic languages Dutch, Yiddish and Afrikaans belong to the former group. Refer-ring to this typological distinction between languages that employ closure voicing and languages that use aspiration to disambiguate two stop series, Iverson & Salmons (2003) introduced the termsvoice language and aspira-tion language, which are adopted in the present thesis.
Regarding phonological representations, there are two main lines of ar-gumentation relating to this typological distinction. One opinion holds that phonology should abstract away from concrete phonetic realisations, em-ploying the same standard feature for all two-way stop contrasts. In the opposite view, features are more directly related to the phonetic realisa-tion of a given phoneme. Consequently, voice languages are assumed to have a different phonological specification than aspiration languages. Hall (2001:31f.) refers to these oppositional approaches as a broad interpretation of the feature [voice] as opposed to a narrow interpretation of the feature.
While the broad interpretation allows [voice] to be an abstract feature to distinguish various kinds of two-way stop contrasts, defenders of the nar-row interpretation allow only those stop series to be specified with a feature [voice] which are regularly produced with vibrating vocal folds. Implemen-tations of a tense/lax contrast that do not make use of closure voicing as a primary cue but rather of ACT are assumed to be specified with a fea-ture other than [voice], for example, [spread glottis] or [tense]. Kageret al.
(2007) refer to a broad interpretation as single feature hypothesis whereas the narrow interpretation is labeledmultiple feature hypothesis, since a single feature [voice] is assumed to be insufficient to describe laryngeal contrasts cross-linguistically. Adopting monovalent features, the respective specifica-tions proposed by the two hypotheses are given in table 3.2.
[b, d, g] [p, t, k] [ph, th, kh] Single Feature voice languages [voice] [ ] — Hypothesis aspiration languages — [voice] [ ] Multiple Feature voice languages [voice] [ ] — Hypothesis aspiration languages — [ ] [spread glottis]
Table 3.2: Feature specifications for two-way laryngeal stop contrasts as sug-gested by the single feature hypothesis and the multiple feature hypothesis.
Instead of [spread glottis], another label might be chosen as a feature for aspi-ration languages.
Since German belongs to the group of aspiration languages, the debate on the abstract representation of laryngeal contrasts has a direct impact on the phonological assumptions made for German stops. According to the single feature hypothesis, the lax series is specified for [voice] and the tense series remains unspecified. The multiple feature hypothesis, in contrast, suggests that the German lax stops /b, d, g/are unspecified while tense /p, t, k/ are specified for a feature relating to aspiration, the most common feature in this constellation being [spread glottis].4
The Single Feature Hypothesis Generally, there are proponents for both approaches. As mentioned before, Wiese (2000) acknowledges the fea-ture framework suggested by Halle & Stevens (1971) which allows for a phonological representation that reflects the phonetic reality rather precisely.
4Note that also binary features could be assumed. The unmarked members of a contrast are represented by [ ] in a privative feature account. In a binary feature system they would be assigned the negative values of the respective feature pair.
3.1 Standard German Stops in Phonology 49 Nonetheless, by continuing to use [±voice] to describe the German laryn-geal system he sticks to the ‘traditional’single feature hypothesis. Another proponent of a broad interpretation is Lombardi (1991, 1995a,b). Kingston
& Diehl (1994) also opt for a [voice]-specification, which they motivate pho-netically. They suggest that the feature [voice] should not only be used to refer to segments that display glottal vibration during closure, but instead to segments that are characterised by the ‘low frequency property’ (see chapter 2, where it is argued that this property is of minor relevance to the Ger-man tense/lax contrast and to laryngeal contrasts in aspiration languages in general). Kingston & Diehl (1994) claim that the low frequency property can be observed in both types of laryngeal contrasts (voiced vs. voiceless and voiceless unaspirated vs. voiceless aspirated) and hence assume that [voice] is the distinctive phonological feature for both voice languages and aspiration languages.
Mikuteit (2006) follows Kingston & Diehl’s (1994) line of argumentation and likewise suggests to specify the German stop contrast with an abstract feature [voice] that may be implemented in two different ways. One possibil-ity is to implement it by a command ‘active vocal fold vibration’, the option chosen by voice languages. In this case, all stops specified for [voice] are produced with closure voicing whereas unspecified stops are marked by the absence of closure voicing. The second possibility is to implement the [voice]
specification by a command ‘no active spreading’, which suppresses a large glottal opening. Thus, Mikuteit (2006) assumes a glottal spreading activity that is turned on by default. Consequently, phonologically unspecified stops are automatically aspirated. Only in segments with a [voice]-specification the active spreading of the glottis is inhibited and the vocal folds are in-structed to adopt a neutral position, resulting in the production of voiceless unaspirated stops. However, intuitively it seems unnatural to assume an extra command to obtain a neutral, relaxed status while the default status consists of an additional activation. Moreover, Mikuteit’s (2006) approach leaves unclear how the assumed default spreading is suppressed in voice languages, where stops are not produced with aspiration.
Another ‘single feature approach’ is proposed by Kohler (1984). Instead of [voice] he suggests to use a ‘power feature’ [±fortis] to express laryngeal contrasts cross-linguistically. He assumes a ‘relative invariance’ between tense and lax stops, which may be realised phonetically differently depending on language and context, but lax stops are always weaker than tense ones.
He argues that [±fortis] is an adequate feature to express phonetic power relations phonologically (with ‘power’ referring to supraglottal movements and the air stream, as well as to tension, especially in the larynx).
The Multiple Feature Hypothesis The opposite opinion, arguing for a multiple feature approach and a narrow interpretation of the feature [voice], has yielded diverse proposals for phonological systems that account more directly for the phonetic properties of phoneme contrasts. For instance, in Keyser & Stevens’s (2006) model on speech processing, [+stiff vocal folds] is the distinctive feature to mark the English tense stop series. Additionally, [spread glottis] is proposed to be a redundant (and language-specific) feature that speakers of English use to enhance the contrast.
Based on a thorough examination of the phonetic properties of German stops, Jessen (1998) proposes to use a feature [±tense] for contrast distinc-tion in German. In his feature selecdistinc-tion process, he closely sticks to articu-latory and acoustic cues found in the German laryngeal contrast. To be able to include Swiss German in the typological group of ‘non-voice languages’, he rejects the feature [spread glottis] which directly refers to aspiration, a phonetic correlate irrelevant for stop distinction in Swiss German (see chap-ter 2). Nonetheless, for Standard German he claims that aspiration is the most reliable phonetic property of the contrast and thus the ‘basic correlate’
of the feature [tense] in this variety of the language.
A strong argument in favour of typologically different phonological speci-fications forvoice languageson the one hand andaspiration languages on the other hand is the assumption that the feature which is active in phonological processes is the marked and specified one. The different phonological be-haviours (e.g., concerning assimilation) found in the two types of languages provide support for the multiple feature hypothesis with its claim that the systems of specification must differ in the respective languages.
Jessen & Ringen (2002) refer to the variation in the production of inter-vocalic lax stops as evidence for a [spread glottis] specification in German.
Since closure voicing cannot be observed consistently, it is – if present – assumed to be passive voicing induced by the adjacent sonorant segments.
Jessen & Ringen (2002:205) claim that the variability is hard to account for if [voice] is assumed to be the active feature, “since it would mean that underlying voiced stops would be optionally devoiced intervocalically, the environment most conducive to voicing”.
Iverson & Salmons (1995; also cf. Iverson & Salmons 1999) suggest that inaspiration languages the tense series of the contrast is specified with the
‘fortis feature’ [spread glottis] while the lax counterparts remain unspeci-fied. The feature [voice], in their view, is the defining feature for two-way contrasts invoice languages, where the tense series remains unspecified. To support their claim, Iverson & Salmons (1995) argue that inaspiration lan-guages there are no phonological processes in favour of a feature [voice].
3.1 Standard German Stops in Phonology 51 Instead, it is rather the voiceless quality that spreads. For example, they refer to sonorant devoicing in obstruent-sonorant clusters in English, where a sonorant following a tense stop becomes voiceless (e.g., ‘plan’ /plæn/˜ → [pl˚æn]). In˜ voice languages like Spanish there is no sonorant devoicing since – as Iverson & Salmons (1995) claim – the voiceless obstruents are not spec-ified for a laryngeal feature and consequently there is no feature to spread its voiceless quality (e.g.,["floriDa]).
A similar, yet structurally different approach is offered by Avery & Id-sardi (2001), who suggest that laryngeal contrasts are not organised by dis-tinctive features but by dimensions which are located at the interface of phonology and phonetics. In their system, the laryngeal node dominates three dimensions, ‘Glottal Width’ (GW), ‘Glottal Tension’ (GT), and ‘Lar-ynx Height’ (LH). The dimensions, in turn, dominate gestures. For each dimension, a universal default completion is assumed, which is the gesture [spread] for GW and [slack] for GT. Hence, the default GW-dimension cor-responds to the feature [spread glottis] as it is applied by Iverson & Salmons (1995, 1999) and the feature [voice] is the equivalent to the default GT-dimension. According to Avery & Idsardi (2001), the phonological contrasts in obstruents are exclusively found on the level of dimensions, the gestures themselves are not contrastive. They are motor instructions, which are added for phonetic realisation. Thus, in their terms, aspiration languages are assumed to specify the tense stop category for GW while the lax one remains unspecified. Voice languages, on the other hand, are represented by a GT/∅ specification, with the lax series being specified and the tense one remaining unspecified. This system was later adopted by Iverson & Salmons (2003) to provide further support for a typological contrast of voice and aspi-ration systems. With the model of dimensional representations they explain the sound shifts from Indo-European to Germanic as a logical change from a GT-system to a GW-system.
Language acquisition data are also used to support the multiple feature hypothesis. Avery & Idsardi (2001:50) state:
“Clearly, phonological inertness must be a primary guide to the child in the setting up of underlying contrasts. If there is no phonological evidence for the presence of a feature through its activity in the phono-logical processes of the language, then there is no reason for the child to utilize the feature in constructing the inventory.”
Kager et al. (2007) show that when acquiring laryngeal contrasts, the er-rors children make in their early productions differ depending on the mother tongue. Dutch children more often erroneously devoice underlyingly voiced
stop-initial targets than vice versa (e.g., /bAl/ is produced as [pAl] instead of [bAl] ‘ball’). German children’s errors are more often of the kind that underlyingly tense (i.e., aspirated) word-initial stops are lenited to voice-less unaspirated stops on the surface (e.g., /"pUp@/ ["phUph@] surfaces as ["pUpa] ‘doll’).5 Following Jakobson’s (1941/72) hypothesis that children’s initial errors mostly consist of a neutralisation to the unmarked values of the phonological features, Kageret al.(2007) argue that under themultiple feature hypothesis both devoicing in Dutch and lenition in German can be accounted for by the absence of the respective laryngeal feature. Dutch chil-dren skip [voice] and thus produce voiceless stops. German chilchil-dren, on the other hand, omit the feature [spread glottis] and thus produce unaspirated stops. Kager et al.(2007) corroborate their claim by showing that English children’s early productions often display laryngeal harmony processes that can be explained best by assuming a feature [spread glottis] for English.
3.1.2 Summary
The preceding paragraphs have shown that there are many different ap-proaches focusing on the phonological representation of laryngeal contrasts.
Without a universally accepted general feature framework, it is clear that there is no unified representational system for the German tense/lax contrast either. The present work does not aim to contribute to the discussion how to represent laryngeal features and, thus, leaves it to others to maybe some day settle the issue. Here, the focus is clearly put on the phonetic imple-mentation and on the acoustic properties of stops, with phonology playing only a subordinate role. What should be kept in mind is that the assumed phonological representation for laryngeal contrasts like the tense/lax opposi-tion in German is based on distinctive features and that the most important phonetic correlate for the German contrast is ACT.